ResearchEnvBench - CUDA Alignment: leaderboard
Metric: Success rate (%) of the cuda-check probe (the installed framework sees the GPU, C2) on the 44 applicable repositories, ResearchEnvBench's 44 post-2024 ML research repositories: the agent starts from a bare Ubuntu container with CUDA 12.4 drivers and two RTX 4090 GPUs, sets up the environment with shell and file-editing tools under a time budget, and hidden probes then check the frozen environment; repositories where the stage does not apply are left out of the denominator; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.5 | 93.2 | #138 |
| 2 | GLM-4.7 | 90.9 | #185 |
| 3 | GPT-5.1 Codex | 79.5 | #154 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=researchenvbench-cuda-alignment · How It Works · Data refreshed daily, snapshot 2026-10-11.